Crusoe
Last reviewed
Jun 9, 2026
Sources
33 citations
Review status
Source-backed
Revision
v2 ยท 5,539 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 9, 2026
Sources
33 citations
Review status
Source-backed
Revision
v2 ยท 5,539 words
Add missing citations, update stale details, or suggest a clearer explanation.
Crusoe is an American AI infrastructure company headquartered in Denver, Colorado, that builds and operates data centers for artificial intelligence workloads. Founded in 2018 by Chase Lochmiller and Cully Cavness, the company began by capturing flared natural gas at oil wellsites to power Bitcoin mining operations before pivoting to hyperscale AI cloud infrastructure in 2023 and 2024.[1][2] As of 2025, Crusoe is best known as the primary developer of the flagship Stargate Initiative data center campus in Abilene, Texas, a 1.2-gigawatt facility built for OpenAI and Oracle.[7] The company raised a $1.375 billion Series E round in October 2025 at a valuation exceeding $10 billion.[3]
Crusoe operates across three business lines: Crusoe Cloud, a GPU rental platform; large-scale infrastructure leasing to hyperscale tenants; and Crusoe Industries, an in-house manufacturing division producing modular data centers and electrical components.[1] The company markets itself as an energy-first AI infrastructure provider, citing its origins in waste-gas recapture and its ongoing focus on renewable and stranded-energy power sources.
Chase Lochmiller and Cully Cavness had been high school classmates before going their separate ways through college and early careers. On a hiking trip in 2017, the two reconnected and began discussing a problem that had bothered Lochmiller during his time at Stanford and in quantitative finance: the enormous amount of natural gas burned off (flared) at oil and gas extraction sites in the United States and globally.
Oil production often releases associated natural gas as a byproduct. In remote locations without pipeline access, operators have historically flared this gas, burning it off as a waste product. The practice contributes to greenhouse gas emissions and represents a significant economic loss. Lochmiller and Cavness saw an opportunity: deploy containerized data centers directly at these wellsites, capture the waste gas, generate electricity with it on-site, and use that electricity for computing workloads. There would be no need for grid transmission infrastructure, and the energy cost would be near zero since the gas would otherwise be destroyed.
They founded Crusoe Energy Systems in Denver in 2018.[2] The name comes from Daniel Defoe's Robinson Crusoe, the castaway who survived by resourcefully using what was available to him on an otherwise abandoned island.
Crusoe's core early technology was its patented Digital Flare Mitigation (DFM) system. The system works by deploying generator equipment and modular data centers directly to oil wellsites. Natural gas that would otherwise be flared is instead fed into Waukesha engine packages, which convert it to electricity through an internal combustion process. That electricity then powers computing hardware housed in trailer-sized modular units.
The environmental case for DFM rested on combustion efficiency. Flares typically achieve around 91% combustion efficiency, meaning a meaningful fraction of the methane in flared gas escapes directly into the atmosphere. Methane is a potent greenhouse gas, roughly 80 times more powerful than carbon dioxide over a 20-year timeframe. Crusoe's engine packages achieved 99.9% combustion efficiency, converting nearly all the methane to carbon dioxide. This significantly reduced the greenhouse gas impact of the associated gas. The company claimed its systems reduced carbon dioxide equivalent emissions by up to 68.6% compared to flaring.[2] Over the life of the DFM business, from 2018 to 2025, Crusoe reported that it repurposed more than 21 billion cubic feet of stranded natural gas, generated about 2.5 terawatt-hours of electricity, and avoided roughly 2.7 million metric tons of carbon dioxide emissions.[2]
Initially, Crusoe used the computing capacity at these wellsite units for Bitcoin mining, which required relatively little management once deployed. Bitcoin mining was well-suited to remote, intermittent power sources because it had no latency requirements and could be paused without consequence when energy was unavailable.
Crusoe formed its manufacturing partnership with Easter-Owens Electric Co., a third-generation family business in the Denver area that built modular data centers and specialized electrical systems. Easter-Owens became a key supplier for Crusoe's DFM units.
Crusoe raised a $4.5 million seed round in March 2019.[18] By December 2019, it had closed a $30 million Series A led by Bain Capital Ventures.[18] As the DFM business grew and the company began expanding to new oil basin regions including North Dakota's Bakken formation, Wyoming, Colorado, and the Permian Basin in Texas, a $128 million Series B led by Valor Equity Partners followed in 2021.[31] Clients including XCL Resources began contracting with Crusoe to handle associated gas at their production sites.
In 2022, Crusoe acquired Easter-Owens outright, vertically integrating its data center manufacturing capability.[9] About 70 Easter-Owens employees joined Crusoe. The acquisition gave Crusoe direct control over the manufacturing of electrical enclosures, switchgear, and modular data center units, reducing dependence on outside suppliers and shortening lead times. Crusoe continued to operate Easter-Owens' facilities, which totaled about 87,000 square feet, and rebranded the division as Crusoe Industries over the following years.[2]
A $350 million Series C led by G2 Venture Partners arrived in 2022 at a valuation of approximately $1.75 billion.[32] Combined with project-level credit facilities, the financing announced at the time totaled about $505 million in new capital and was used in part to prepare the launch of Crusoe Cloud.[32]
By 2022 and early 2023, Crusoe was operating a large fleet of DFM units across multiple oil basins, primarily running Bitcoin mining workloads. Two factors pushed management toward a significant strategic shift.
First, the market for AI computing exploded following OpenAI's release of ChatGPT in late 2022 and the subsequent explosion in demand for GPU training and inference capacity. Demand for NVIDIA H100 GPUs far outstripped supply. Cloud providers and AI labs were scrambling for compute.
Second, Bitcoin's volatility and the structural pressure on mining margins made it a less attractive long-term revenue base. The company had the operational expertise and energy procurement skills to serve AI workloads, which were far more lucrative and had stronger long-term demand tailwinds.
In 2023, Crusoe extended Crusoe Cloud, its GPU rental platform, to support general HPC workloads including AI training and inference.[5] The company secured a $200 million loan from Upper90 in late 2023 with stated plans to use the funds to acquire NVIDIA H100 GPUs at scale.[2]
A $600 million Series D followed in early 2024, led by Founders Fund at a valuation of approximately $2.8 billion.[2] Investors in that round included Fidelity Investments, NVIDIA, Lowercarbon Capital, Winklevoss Capital, DRW Trading Group, Mubadala, Valor Equity Partners, and Ribbit Capital. NVIDIA's participation was notable: it signaled that Crusoe had a preferred position in the queue for GPU allocation at a time when H100 supply was severely constrained.
Crusoe's most consequential strategic move came in mid-2024. The company announced it would develop a large-scale data center campus at the Lancium Clean Campus in Abilene, Texas. Lancium operates a renewable-energy-optimized industrial campus in West Texas with access to low-cost power from the Texas wind corridor.
The initial announcement described a 200-megawatt AI data center. Shortly afterward, Oracle signed a 15-year lease agreement with Crusoe at the site, with the first contract covering approximately 220 megawatts of IT capacity. Oracle's involvement connected the Abilene project to the broader Stargate Initiative, the large-scale AI infrastructure buildout announced by OpenAI, Oracle, SoftBank, and MGX in January 2025 at a White House press conference alongside President Donald Trump. The Stargate initiative committed to $500 billion in AI infrastructure investment over four years.
The Abilene campus became Stargate's flagship site. By March 2025, Crusoe expanded plans for the campus to a total of 1.2 gigawatts across eight buildings totaling approximately four million square feet, with the first two buildings (about 980,000 square feet, 200+ megawatts) going live in September 2025.[7][8] Oracle subsequently signed a second contract in early 2025 covering an additional 660 megawatts of capacity, bringing the total contract value for the Oracle-Crusoe relationship to an estimated $15 to $20 billion.
In October 2024, Crusoe, Blue Owl Capital, and Primary Digital Infrastructure announced a $3.4 billion joint venture to fund the initial phase of the Abilene development.[10] A second phase of the joint venture followed in early 2025, growing the total financing commitment to $15 billion.
In March 2025, Crusoe announced it had agreed to sell its entire Bitcoin mining and DFM business to NYDIG, a Bitcoin financial services firm.[4] The transaction included Crusoe's fleet of over 425 modular DFM data centers, its power generation technology representing over 270 megawatts of capacity, and approximately 135 Crusoe employees who continued operating the business under NYDIG.[16] Crusoe retained an equity stake in the combined entity.
The sale completed Crusoe's transition. The Bitcoin mining segment had comprised roughly 55% of Crusoe's 2024 revenue, contributing about $152 million.[2] Its elimination removed cryptocurrency market exposure and allowed the company to focus entirely on AI infrastructure. After the deal closed, Crusoe described itself as "the AI factory company."[3]
In October 2025, Crusoe announced the initial closing of a $1.375 billion Series E round, co-led by Valor Equity Partners and Mubadala Capital.[3] The round valued the company at more than $10 billion.[3] Participants included NVIDIA, Fidelity Management and Research, T. Rowe Price, Tiger Global Management, Salesforce Ventures, Founders Fund, Spark Capital, Altimeter Capital, Franklin Templeton, Winklevoss Capital, and Supermicro, among others.[3] Additional named investors included StepStone Group, Ribbit Capital, Lowercarbon Capital, Radical Ventures, 137 Ventures, and MCJ, with Blue Owl expected to join in a later closing.[3]
The company reported that Crusoe Cloud bookings grew fivefold in the first three quarters of 2025 compared to the same period in 2024.[3] Projected 2025 revenue was approximately $998 million, representing roughly 262% year-over-year growth.[2] Total equity and debt raised reached approximately $3.9 billion by the time of the Series E announcement.[3] Crusoe also reported that its power pipeline had grown more than fourfold to over 45 gigawatts.[3]
In December 2025, Crusoe won "North American Data Center Project of the Year" at the Data Center Dynamics Global Awards, recognizing the Abilene campus.[17]
In March 2026, Axios reported that Crusoe had begun raising a pre-IPO funding round of up to $2 billion at a valuation of $30 billion to $40 billion, a sharp step up from the $10 billion valuation set in October 2025.[21] The company has been widely viewed as a candidate for an initial public offering in 2026 as it races to meet demand for AI computing and energy capacity.[21]
In February 2026, Crusoe arranged a $300 million loan from Goldman Sachs to purchase AMD AI accelerators for a data center in Ohio developed by 5C.[28] AMD agreed to guarantee the loan and to lease the chips back if customer demand fell short, an arrangement reported as the first known instance of AMD using such financing to support a chip sale, similar to the circular deals NVIDIA had made.[28] The loan carried a roughly 6% interest rate and was secured by the chips and equipment.[28] CEO Chase Lochmiller said the company planned to buy about $400 million in new AMD chips and expected to spend roughly $4 billion annually over the following decade on expansion.[28]
Chase Lochmiller studied mathematics and physics as an undergraduate at MIT and completed a master's degree in computer science at Stanford, where he focused on artificial intelligence. After Stanford, he worked as a quantitative researcher and algorithmic trader at GETCO and Jump Trading, two Chicago-based high-frequency trading firms. His background in quantitative methods and his exposure to large-scale computing infrastructure during the cryptocurrency boom informed the technical and financial architecture of Crusoe.
Lochmiller has served as CEO since the company's founding. He has spoken publicly about the company's thesis that the most durable advantage in AI infrastructure is not GPU procurement but energy: access to power at favorable cost and at the scale required by modern AI workloads.
Cully Cavness graduated from Middlebury College in 2010 with a degree in geology. Middlebury has a strong emphasis on environmental sustainability, and Cavness's geological background gave him a working understanding of the oil and gas industry's extraction processes and the flaring problem that underpinned Crusoe's original business. After college, he worked in energy investment banking before the 2017 hiking trip where the Crusoe concept took shape.
Cavness has served as President of Crusoe since the company's founding, overseeing operations and business development. He presented testimony to North Dakota legislative committees in 2019 and 2021 explaining the DFM technology and its regulatory context. As the company scaled in 2025 and 2026, Cavness's title was given as President and Chief Strategy Officer, and Crusoe appointed Michael Gordon as Chief Operating Officer and Chief Financial Officer.[33]
| Round | Date | Amount | Lead Investor | Valuation |
|---|---|---|---|---|
| Seed | March 2019 | $4.5M | Various | Undisclosed |
| Series A | December 2019 | $30M | Bain Capital Ventures | Undisclosed |
| Series B | 2021 | $128M | Valor Equity Partners | Undisclosed |
| Series C | 2022 | $350M | G2 Venture Partners | ~$1.75B |
| Series D | 2024 | $600M | Founders Fund | ~$2.8B |
| Series E | October 2025 | $1.375B | Valor Equity Partners, Mubadala Capital | >$10B |
| Pre-IPO (in progress) | 2026 | up to $2B | Undisclosed | $30B-$40B (reported) |
In addition to equity rounds, Crusoe has raised significant debt financing: a $200 million loan from Upper90 in late 2023, a $750 million credit facility from Brookfield in June 2025,[24] a $175 million credit facility from Victory Park Capital for the Iceland expansion, a $300 million AMD-guaranteed loan from Goldman Sachs in February 2026,[28] and a J.P. Morgan-led construction loan reported at about $7.1 billion for the Abilene campus.[33] Total equity and debt financing exceeded $4.4 billion by early 2026, before the project-level construction debt is counted.
Crusoe Cloud is the company's GPU rental service, offering on-demand, spot, and reserved compute capacity to AI researchers, developers, and enterprises. The platform supports NVIDIA and AMD GPU hardware across a range of configurations, from single-GPU deployments to large multi-node clusters interconnected via InfiniBand networking.
As of early 2026, Crusoe Cloud offers the following GPU configurations:
| GPU | VRAM | On-demand price (per GPU-hour) | Spot price |
|---|---|---|---|
| NVIDIA GB200 NVL72 | 186GB | Contact sales | Contact sales |
| NVIDIA B200 HGX | 180GB | Contact sales | Contact sales |
| NVIDIA H200 HGX | 141GB | $4.29 | Contact sales |
| NVIDIA H100 HGX | 80GB | $3.90 | Varies |
| AMD MI355X | 288GB | Contact sales | Contact sales |
| AMD MI300X | 192GB | $3.45 | Varies |
| NVIDIA A100 SXM | 80GB | $1.95 | ~$1.30 |
| NVIDIA A100 PCIe | 80GB | $1.65 | ~$1.20 |
| NVIDIA L40S | 48GB | $1.00 | ~$0.50 |
| NVIDIA A40 | 48GB | $0.90 | ~$0.40 |
Billing is per-minute with no upfront setup fees.[20] The company charges no data egress fees, a meaningful differentiator for customers running workloads that generate large volumes of output data.
Crusoe Cloud uses lightweight virtual machines rather than bare metal, providing rapid provisioning, stateful snapshots, shared storage, and network isolation. The platform is designed to support large distributed training jobs requiring tightly coupled GPU-to-GPU communication, with InfiniBand networking enabling the low-latency interconnects that large language model training requires.
Crusoe offers managed services including storage, networking, and Kubernetes orchestration. The company received a SOC 2 Type II certification for its cloud operations.
In March 2025, Crusoe Cloud was rated "Gold" in SemiAnalysis's ClusterMAX GPU cloud rating system, a tier shared with providers such as Nebius, Oracle, Azure, and Together AI, with the report singling out Crusoe's console interface and automated fault management for praise.[29] CoreWeave was the only provider rated in the top "Platinum" tier at that time.[29]
Customer reviews have noted high cluster uptime; Windsurf, a developer tools company, reported 99.98% uptime on its H100 cluster running on Crusoe.
The Abilene campus is Crusoe's largest and most strategically significant site. Located at the Lancium Clean Campus in Taylor County, Texas, the campus benefits from Lancium's proprietary energy management system designed to access low-cost power from the Texas wind corridor.
The campus currently consists of the first two buildings (totaling approximately 980,000 square feet and 200+ megawatts of IT capacity), which went live in September 2025.[6] Six additional buildings are under construction with completion expected in mid-2026, bringing the campus to approximately four million square feet and 1.2 gigawatts of total power capacity.[7] Each building is engineered to operate up to 50,000 NVIDIA GB200 NVL72 systems on a single integrated network fabric.
The site is designed for direct-to-chip liquid cooling and can also accommodate air cooling. Approximately 2,000 construction workers were on-site daily during the initial build phase, with the workforce expected to grow to 5,000 during the expansion phase.
Oracle has a 15-year lease for the facility, and the campus serves as OpenAI's primary compute site under the Stargate project. In early 2026, Microsoft announced it was constructing two "AI factory" buildings and an on-site power plant adjacent to the Crusoe campus.[19] In March 2026, Crusoe formally announced this as a separate 900-megawatt AI factory campus consisting of two buildings (each carrying about 336 megawatts of critical IT load) with a dedicated 900-megawatt on-site power plant, bringing the Abilene complex to 10 buildings and approximately 2.1 gigawatts in total.[22] The first Microsoft building is expected to be energized in mid-2027.[22] Microsoft picked up this capacity after OpenAI and Oracle declined to expand their own footprint further; OpenAI said it "considered expanding it further, but ultimately chose to put that additional capacity in other locations."[19]
Crusoe partnered with atNorth to expand cloud capacity at atNorth's ICE02 data center in Iceland.[13] The facility is powered entirely by geothermal and hydroelectric energy, both of which are abundant in Iceland. The Iceland deployment focuses on NVIDIA Blackwell hardware, including NVIDIA DGX GB200 NVL72 instances. A $175 million credit facility from Victory Park Capital supports the Iceland expansion.
In July 2025, Crusoe announced a partnership with Tallgrass Energy Partners to develop a 1.8-gigawatt AI data center campus in Laramie County, southeast Wyoming.[14] The campus is expandable to 10 gigawatts in later phases.[14] Laramie County commissioners voted unanimously on January 6, 2026 to approve construction of the data center, known as Project Jade, together with Tallgrass's adjacent gas-fired power facility.[30] On-site power generation will use natural gas turbines operated by Tallgrass, with carbon capture facilitated by proximity to Tallgrass's existing CO2 sequestration infrastructure. The combined data center and power project was reported to scale toward about 2.7 gigawatts, with the energy infrastructure alone representing roughly $7 billion of investment and total capital expenditure expected to exceed $50 billion.[30] Construction on the first buildings is expected to complete in 2027.[14]
In early 2026, Crusoe entered a multi-year framework agreement with Kalina Distributed Power to develop multiple AI data centers in Alberta, Canada, each co-sited with a 170-megawatt natural gas power plant using carbon capture and sequestration.[25] Crusoe is to develop, own, and operate the data centers and purchase power under power purchase agreements with a minimum 15-year term, across sites including the Myers, Alsike, and Crossfield energy parks.[25]
In April 2026, Crusoe agreed to lease about 40 megawatts of capacity in a data center being developed in Afula, Israel, by the Anan Group, in a deal valued at about NIS 3 billion over a decade.[26] Crusoe was reported to plan the purchase of roughly 12,000 NVIDIA processors for the leased space.[26]
Crusoe Cloud also serves customers through third-party data centers in Virginia and Montana. The company has announced plans for additional sites in Norway as part of a broader European expansion toward renewable energy sources.
Crusoe Industries is the company's in-house manufacturing and engineering division, created through the 2022 acquisition of Easter-Owens Electric.[9] The division operates facilities in Arvada, Colorado (near Denver), Tulsa, Oklahoma, and Ponchatoula, Louisiana.
Crusoe Industries manufactures a range of components used in Crusoe's own data centers as well as for external clients: metal structures, electrical enclosures, industrial control systems, power distribution equipment, and switchgear. A key competitive argument is that by producing switchgear internally, Crusoe reduced lead times from roughly 100 weeks (the typical external supplier timeline during the post-pandemic supply chain shortage) to approximately 22 weeks.[2] Faster electrical component delivery enables Crusoe to commission data centers more quickly than competitors relying on outside vendors.
In June 2025, Crusoe launched Crusoe Spark, a line of prefabricated, turnkey modular AI data centers designed for edge deployments.[12] Spark units can be delivered and operational within three months, supporting on-premises AI inference, edge computing, and distributed AI workloads. Power inputs are flexible, accommodating grid power, on-site generation, or stranded energy sources.
In March 2026, Crusoe announced the Spark Factory, a dedicated manufacturing facility in Brighton, Colorado, covering 352,000 square feet, specifically for producing Crusoe Spark modular units.[27] The company described the combination of the factory and an initial fleet of Spark units as a commitment of more than $200 million, expected to create over 200 local jobs, with the first factory-produced modules due in the third quarter of 2026.[27]
The Stargate Initiative is a joint venture announced in January 2025 by OpenAI, Oracle, SoftBank, and MGX to build AI data center infrastructure in the United States. The initiative committed up to $500 billion over four years. The Abilene campus built by Crusoe was the first Stargate site to go live.[6]
Crusoe's role at Abilene is that of developer and operator. Oracle holds the long-term tenant relationship and provides compute to OpenAI. The arrangement put Crusoe at the center of the most high-profile AI infrastructure project in the world. However, the relationship faced some difficulty: a multi-day outage at the Abilene campus caused by winter weather affecting liquid cooling equipment reportedly damaged relations between OpenAI and Crusoe in early 2026, and Oracle and OpenAI subsequently scaled back expansion plans for the flagship site.[11] Reports in early 2026 indicated that Meta was in discussions with Crusoe and NVIDIA about acquiring some of the capacity that Oracle and OpenAI had declined to take.[11] According to those reports, the winter storm left insufficient glycol concentration in the chiller loops, ice blocked cooling headers, and emergency shutdowns extended downtime; the episode was also linked to the departure of Crusoe's director of physical infrastructure.[11] NVIDIA was reported to have paid a deposit of about $150 million to secure the contested space before approaching Meta about taking the lease.[11]
Despite the complications, Crusoe remained one of the largest data center developers in the Stargate ecosystem. Microsoft moved forward on the adjacent 900-megawatt construction at the Abilene site, taking capacity that OpenAI declined.[19][22]
Crusoe's energy strategy has changed substantially since the company's founding, but the underlying thesis has remained consistent: the cost and source of energy is a structural differentiator in data center economics.
In its original DFM business, Crusoe captured natural gas that would otherwise be flared, converting a waste product into computing electricity. The environmental benefit was real, because burning methane in an efficient engine produces far less climate impact than allowing it to escape or burning it inefficiently in a flare.
After selling the DFM and Bitcoin mining operations to NYDIG, Crusoe's sustainability focus shifted toward renewable power sourcing for its large-scale data centers. The Iceland facility runs on 100% geothermal and hydroelectric power. The Lancium Clean Campus in Abilene is connected to West Texas wind generation, and Lancium's energy management software is specifically designed to maximize consumption of low-cost renewable power while maintaining stable operations.
The Wyoming campus introduces a more complicated picture: power will come from Tallgrass natural gas turbines rather than renewables, with carbon capture infrastructure nearby.[14] Crusoe has described this as a transition-period approach while longer-term renewable development at the site moves forward. The company has further extended this gas-plus-carbon-capture model to Alberta, where each planned data center is paired with a 170-megawatt near-zero-emissions gas plant.[25] In March 2025, Crusoe and the investment firm Engine No. 1 formed a joint venture to secure up to 4.5 gigawatts of behind-the-meter natural gas power for AI campuses, drawing on GE Vernova turbines that Engine No. 1 had acquired with Chevron, with delivery expected around 2027.[23]
Crusoe has received attention from ESG-focused investors, and climate-focused funds including Lowercarbon Capital and MCJ participated in its Series E.[3] The company received the 2025 Data Center Dynamics "North American Data Center Project of the Year" award partly on the basis of its sustainability approach at Abilene.[17]
Crusoe occupies a position in the GPU cloud market sometimes described as a "neocloud," a category of infrastructure providers that emerged alongside the AI boom and focus specifically on GPU compute for AI workloads, as opposed to the general-purpose public clouds (AWS, Azure, Google Cloud) that hold roughly 63% of the broader cloud market.
| Provider | Founded | GPU focus | Energy approach | Scale | Public/Private | Key differentiator |
|---|---|---|---|---|---|---|
| CoreWeave | 2017 | NVIDIA H100/H200/GB200 | Grid-dependent, some renewable sourcing | ~$5B+ ARR (2025), IPO March 2025 | Public (NASDAQ: CRWV) | Bare metal, Kubernetes-native, largest neocloud |
| Crusoe | 2018 | NVIDIA H100/H200/B200/GB200, AMD MI300X/MI355X | Renewable, stranded energy, natural gas with carbon capture | ~$1B projected 2025 revenue | Private | Energy-first model, vertical integration, Stargate developer |
| Lambda Labs | 2012 | NVIDIA H100/H200 | Grid-dependent | Smaller scale | Private | Developer-friendly, on-prem + cloud hybrid |
| Nebius | 2023 (reorg) | NVIDIA H100/H200/GB200 | Data center colocation | ~$7-9B ARR guidance 2026 | Public (NASDAQ: NBIS) | ODM hardware, lightweight VMs, competitive pricing |
CoreWeave is the largest neocloud and the most direct competitive comparison for Crusoe. CoreWeave went public in March 2025 and has revenue measured in the billions of dollars. The two companies differ in several meaningful ways.
On architecture, CoreWeave emphasizes bare metal deployments for maximum GPU performance, while Crusoe uses lightweight VMs. CoreWeave bills per-second, which suits bursty interactive workloads, while Crusoe bills per-minute, which is more cost-effective for long batch training runs.
On business model, CoreWeave is primarily a GPU rental platform with large long-term contracts with Microsoft and OpenAI. Crusoe has a broader strategy that combines GPU rental with hyperscale infrastructure development (building and leasing entire data centers) and manufacturing (Crusoe Industries). CoreWeave's revenue base is larger; Crusoe's capital base per unit of revenue is also large because of the scale of its infrastructure investments.
On energy, CoreWeave sources power from grid connections and does not highlight energy sourcing as a primary differentiator. Crusoe's brand identity centers on its energy strategy.
Lambda Labs is smaller than Crusoe and targets developers and research teams with simpler interfaces and competitive pricing. Lambda is well-regarded for its "1-Click Clusters" that simplify multi-node GPU provisioning and for its hybrid model that includes both cloud rentals and on-premises hardware. Lambda does not have the infrastructure development business or the manufacturing capabilities that Crusoe has built.
Nebius is a GPU cloud provider that went public on NASDAQ in 2024 after reorganizing out of Yandex's international assets. Nebius uses ODM hardware and lightweight virtualization similar to Crusoe's VM-based approach, and it rates at the "Gold" tier in independent benchmarks such as SemiAnalysis's ClusterMAX system.[15] Nebius is notable for having secured a large Microsoft contract and a Meta contract, giving it significant hyperscale revenue. In terms of GPU pricing, Nebius has historically been more aggressive than Crusoe at the low end of the market. Crusoe's differentiation relative to Nebius rests more on its domestic US manufacturing, energy sustainability positioning, and role as a data center developer rather than purely a GPU rental company.
Several risks are associated with Crusoe's current business model.
GPU rental rates have declined sharply as supply has caught up with demand. Market prices for H100 compute fell from roughly $8 per GPU-hour in 2023 to around $2 to $4 per GPU-hour by 2025. Continued rate compression would reduce margins on Crusoe Cloud's rental business.
The company carries significant debt obligations. The Abilene campus alone involved billions of dollars in construction financing, and the Wyoming campus will require similar outlays. With projected annual interest expense of around $300 million, revenue growth needs to outpace debt service costs.
Dependency on a small number of large customers creates concentration risk. A substantial portion of Crusoe's infrastructure pipeline is tied to Oracle, OpenAI, and the Stargate initiative. The early 2026 reports about Oracle and OpenAI scaling back Abilene expansion plans illustrate how quickly demand forecasts can shift.[11]
The broader AI infrastructure build-out carries the general risk of overbuilding. If AI workload growth slows or if hyperscalers consolidate demand with fewer providers, smaller neoclouds would face margin pressure.
The winter 2026 cooling outage at Abilene damaged Crusoe's operational reputation with a key customer and highlighted the reliability risks inherent in novel liquid cooling systems at large scale.[11]
Finally, Crusoe remains a private company as of early 2026. While its valuation has risen sharply, it has not yet achieved the liquidity event that its investors are presumably expecting, and comparisons to CoreWeave's post-IPO market cap are difficult to make directly. As of March 2026 the company was reported to be raising a pre-IPO round and was widely viewed as a candidate for a public listing during 2026.[21]